In this chapter, we're going to build a deep Q-network to solve the well-known CartPole (inverted pendulum) problem. We'll be working with the OpenAI Gym CartPole-v1 environment. We'll also use Keras with a TensorFlow backend to implement our deep Q-network architecture.
We'll become familiar with OpenAI Gym's CartPole-v1 task and design a basic Deep Q Learning (DQN) structure. We'll construct our deep learning architecture using Keras and start to tune the learning parameters and add in epsilon decay to optimize the model. We'll also add in experience replay to improve our performance. At each iteration of our model-building process, we'll run a new training loop to observe the updated results.
The following topics will be covered in this chapter:
- Getting started with the CartPole...
